Preference Reversals in Judgment and Choice: The Prominence Effect
Marcus Selart
Department of Psychology
Göteborg University
Preference Reversals in Judgment and Choice:
The Prominence Effect
Avhandling för filosofie doktorsexamen i psykologi, som med vederbörligt tillstånd av Samhällsvetenskapliga fakulteten vid Göteborgs universitet kommer att offentligen försvaras onsdagen den 23 november 1994, kl. 10.00 i sal L 5, Psykologiska institutionen, Skårs Led 3, Göteborg.
av
Marcus Selart
Doctoral dissertation at Göteborg University 1994
Abstract
Selart, M., 1994. Preference reversals in judgment and choice: The prominence effect. Department of Psychology, Göteborg University, Sweden. ISRN GU/PSYK/AVH--12--SE
According to normative decision theory there exists a principle of procedure invariance which states that a decision maker's preference order should remain the same, independently of which response mode is used. For example, the decision maker should express the same preference independently of whether he or she has to judge or decide. Nevertheless, previous research in behavioral decision making has suggested that judgments and choices yield different preference orders in both the risky and the riskless domain. In the latter, the prominence effect has been demonstrated.
The main purpose of the present series of experiments was to test cognitive explanations which account for the prominence effect. One of the explanations provided a psychological account based primarily on decision-strategy compatibility. Two other explanations built on information structuring approaches. In the first one, the general idea was that decision makers differentiate between alternatives by value and beiief restructuring. In the second approach, violations of invariance were assumed to be attributed to the information structure of the task which in many cases demand problem simplification.
A prominence effect was in most experiments found for both choices and preference ratings.
This finding spöke against the strategy compatibility explanation. Instead, the different forms of cognitive restructuring provided a better account. However, none of these provided a single explanation. Yet, the structure compatibility explanation appeared to be the more viable one, in particular of the relation between experimentäl manipulations and response mode outcomes. The predictions of the value-belief restructuring explanation, on the other hand, seemed to be more valid for the prominence effect found in choice than for preference ratings.
Belief-value - Compatibility - Decision making - Information processing - Preference reversals - Prominence effect - Response mode - Restructuring
Marcus Selart, Department of Psychology, Göteborg University, PO Box 14158, S-400 20 Göteborg, Sweden. Email: Marcus.Selart@psy.gu.se
ISSN 1101-718X ISRN GU/PS YK/AVH-- 12--SE
Preference Reversals in Judgment and Choice: The Prominence Effect
Marcus Selart
Department of Psychology Göteborg University
Sweden
Printed in Sweden
Vasastadens Bokbinderi AB 1994
ISSN 1101-718X ISRN GU/PSYK/AVH—12—SE
To my family
Doctoral dissertation at Göteborg University 1994
Abstract
Selart, M., 1994. Preference reversals in judgment and choice: The prominence effect. Department of Psychology, Göteborg University, Sweden. ISRN GU/PSYK/AVH--12--SE
According to normative decision theory there exists a principle of procedure invariance which states that a decision maker's preference order should remain the same, independently of which response mode is used. For example, the decision maker should express the same preference independently of whether he or she has to judge or decide. Nevertheiess, previous research in behavioral decision making has suggested that judgments and choices yield different preference orders in both the risky and the riskless domain. In the latter, the prominence effect has been demonstrated.
The main purpose of the present series of experiments was to test cognitive explanations which account for the prominence effect. One of the explanations provided a psychological account based primarily on decision-strategy compatibility. Two other explanations built on information structuring approaches. In the first one, the general idea was that decision makers differentiate between altematives by value and belief restructuring. In the second approach, violations of invariance were assumed to be attributed to the information structure of the task which in many cases demand problem simplification.
A prominence effect was in most experiments found for both choices and preference ratings.
This finding spöke against the strategy compatibility explanation. Instead, the different forms of cognitive restructuring provided a better account. However, none of these provided a single explanation. Yet, the structure compatibility explanation appeared to be the more viable one, in particular of the relation between experimental manipulations and response mode outcomes. The predictions of the value-belief restructuring explanation, on the other hand, seemed to be more valid for the prominence effect found in choice than for preference ratings.
Belief-value - Compatibility - Decision making - Information processing - Preference reversals - Prominence effect - Response mode - Restructuring
Marcus Selart, Department of Psychology, Göteborg University, PO Box 14158, S-400 20 Göteborg, Sweden. Emaxl: Marcus.Selart@psy.gu.se
ISSN 1101-718X ISRN GU/PSYK/AVH--12-SE
Preface
Acknowledgements
Any thesis that reports psychological research depends on collaboration. This thesis is no exception, and I am grateful for help from several people. My supervisors, Professor Henry Montgomery, Deptartment of Psychology, Stockholm University, and Professor Tommy Gärling, Deptartment of Psychology, Göteborg University, have been encouraging and most helpful during the project. The research was also carried out in collaboration with Docent Erik Lindberg, Swedish Road and Traffic Institute, Linköping, and Mr. Joakim Romanus, Department of Psychology, Göteborg University.
Many people have contributed indirectly to the research. Since it was carried out at the Department of Psychology, Göteborg University, I wish to express my gratitude to the following colleagues: Docent Carl Martin Allwood, Docent Anders Biel, Mr. Ulf Dahlstrand, Mr. Niklas Fransson, Mr. Robert Gillholm, Dr.
Lisbeth Hedelin, Dr. Sven Hemlin, Mr. Bengt Jansson, Mr. Tomas Kalén, Dr.
Gunilla Torell, and Mrs. Helena Willén.
Several people at the department have also read and commented on the entire manuscript, and for this I wish to thank: Professor Trevor Archer, Professor Lars Bäckman, Docent Sven G. Carlsson, Mr. Pär-Anders Granhag, Professor Erland Hjelmqvist, Dr. Bernt Johnsson, Professor Gerry Larsson, and Docent Joseph Schaller.
Others which have provided useful advise and comments are: Dr. Terry Hartig, University of California, Berkeley, Professor Denis Hilton, Groupe ESSEC, Cergy-Pontoise, Professor Oswald Huber, University of Fribourg, Professor Helmut Jungermann, Berlin University of Technology, Dr. Anton Kuehberger, University of Salzburg, Professor Maria Lewicka, University of Warsaw, Dr.
John Maule, Leeds University, Dr. Rob Ranyard, Bolton Institute of Higher Education, Professor Ola Svenson, Stockholm University, Professor Tadeusz Tyszka, Polish Academy of Sciences, Warsaw, Professor Bas Verplanken, University of Nijmegen, Dr. Mirjam Westenberg, University of Amsterdam, and Professor Dan Zakay, Tel-Aviv University.
The research presented in this thesis received financial support from the
Swedish Council for Research in the Humanities and the Social Sciences (HSFR),
the Royal Swedish Academy of Sciences, the Royal Society for Arts and Sciences
in Gothenburg, the Anna Ahrenberg Foundation, the Magnus Bergwall
Foundation, and the Göteborg University Foundations. I especially want to thank
Mrs. Barbro Hänström at the HSFR for excellent administrative support.
Finally, I also wish to express my gratitude to the technical and administrative staff at the Department of Psychology, Göteborg University.
List of publications
The present thesis is based on the following four research papers, which will be referred to in the text by their Roman numerals:
I Montgomery, H., Gärling, T., Lindberg., & Selart, M. (1990). Preference judgments and choice: Is the prominence effect due to information integration or information evaluation ? In K. Borcherding, O. L. Larichev, and D. M. Messick (Eds), Contemporary Issues in Decision Making.
Amsterdam: Elsevier Science Publishers B.V. (North-Holland).
II Montgomery, H., Selart, M., Gärling, T., & Lindberg, E. (1994). The judgment-choice discrepancy: Noncompatibility or restructuring? Journal of Behavioral Decision Making, 7, 145-155.
III Selart, M., Montgomery, H., Romanus, J., & Gärling, T. (1994). Violations of procedure invariance in preference measurement: Cognitive explanations.
European Journal of Cognitive Psychology. In Press.
IV Selart, M. (1994). Can accountability and value differences modify the prominence effect in judgment and choice? Göteborg Psychological Reports, 24, 1-19.
October 1994,
Marcus Selart
Contents
1.0 Introduction
1.1 Perspectives on rationality
1.2 General introduction to the present studies
2.0 Violations of Invariance 2.1 Assumptions of invariance 2.2 Distortions in attribute weighting 2.3 Cost-benefit theories
2.4 Restructuring theories
3.0 Summary of the Empirical Studies 3.1 Study I
3.2 Study II 3.3 Study III 3.4 Study IV
4.0 General Discussion and Conclusions References
Appendix
1.0 Introduction
1.1 Perspectives on rationality
In the behavioral sciences and in other disciplines, the area of decision making research has been given considerable attention. The main reason for this is that how we decide is a crucial phenomenon for the functioning of social systems. In fact, it can be argued that to what degree we are able to understand and interpret a social system depends on our knowledge of how people make effective decisions.
In the discipline of psychology research has also to a high extent been concentrated on fundamental characteristics of basic mental functions that humans share with animals. Such processes are the acquisition of information from the environment, emotional reactions, learning, and memory. It is important to emphasize that no conflicts exist between the basic assumptions of these two directions within psychology. Higher mental processes like decision making has the potential of being linked to physiology. The connections are simply somewhat looser than to the basic mental processes (Campbell, 1986; Rachlin, 1989).
At the same time the contribution of psychological research in decision making to other social sciences, especially economics, is essential. Many economists today realize that issues like cognitions, emotions, and behavior must be taken into account in their rational models (Lucas, 1986; Plott, 1986). Still, there exist some general differences between economics and psychology in the study of human decision making. First, whereas economists focus on outcomes, questions of process are central to psychologists. Thus, psychologists are both concerned with the manner in which decisions are made as well as with the characteristics of the participants and of their resource endowments. Second, the empirical evidence in psychology is experimental in nature. Much work in modern economics is theoretical, that is, many economists are more interested in deriving implications from theory than performing experiments and tests. Nevertheless, there is a growing interest for "experimental economics" among economists. This tendency is perhaps most apparent in areas such as marketing, accounting, and public policy.
Over the years most investigations in the field of behavioral decision making
have more or less explicitly dealt with the notion of rationality. On the one hand
researchers have argued that all sorts of biases, errors, and inaccuracies speak
against that people are generally rational. On the other hand, when faced with the
suggestion that a person is irrational, the same researchers might have argued
against this and in favör of the good reasons of that person. Of course, issues like what is valid in general and exceptions from what is thought of as common can explain such different approaches which scientists have, but the rationality debate indicates underlying opposing perspectives (Berkeley & Humphreys, 1982; Cohen, 1979; Edwards, 1983; Einhorn & Hogarth, 1981; Fischhoff, 1983; Kahneman &
Tversky, 1982; Nisbett & Ross, 1980; Phillips, 1983).
The rationality concept has been developed in philosophy and economics, and therefore these disciplines have had a great impact on the definition of the concept in psychology. The common assumption in behavioral sciences that people are expected to act in line with their values and beliefs, can be seen as an expression of this. A rational choice can according to Dawes (1988) be defined as one that meets three criteria: (i) It is based on the current assets of the decision maker. Assets include not only money, but physiological state, psychological capacities, social relationships, and feelings, that is vital phenomena for our well-being from a psychosomatic point of view. (ii) It is based on the possible outcomes of the choice. (iii) When these outcomes are uncertain, their likelihood is evaluated without violating the basic rules of probability theory. Such definitions have as their aim to satisfy the claims from philosophy that a rational action is "logical" or
"consistent" as stated in a set of axioms. Rationality is hence specified normatively, that is, the mission of behavioral research is often to study empirically whether actual human behavior is rational in the sense that it obeys the norm. According to Dawes (1988) there are common decision making procedures that have no relationship to the criteria of rationality. They involve habit, tradition, conformity, convictions, and imitation.
However, in modern cognitive psychology the use of a normative yardstick is
becoming more and more rare. Nevertheless, there are exceptions. For instance,
the physical environment is often regarded as a standard in the study of perceptual
illusions. But this physical standard is usually not treated as a yardstick. Instead
physical stimuli are used to trigger responses which are not regarded as "deviant
from" or "consistent with" the stimuli being used. Another exception is the
assumption of an ideal speaker or listener which was formerly widely used in
psycho-linguistic research. The growing use of semantic network theories has
however turned the focus of interest towards the understanding of actual human
use of language. Yet another exception is that in research on thinking and
reasoning formål logic is still used as a norm, but as in the case of psycho-
linguistics, the interest is slowly turning towards the content of thinking in place of
its form.
In behavioral decision making research, normative models have, on the other hand, been heavily used since the early fifties. Research has had as its major objective to study, explain, and interpret discrepancies between predictions based on normative theories and actual decision/judgmental behavior. In the study of decisions the subjective expected utility (SEU) model represents the normative model which has been used whereas the most commonly used models of judgment are Bayes' theorem and multiattribute utility (MAUT) models. Much debate in the area still concerns these models, and the assumption agreed upon is the idea of rationality embodied in these models.
Briefly, the subjective expected utility (SEU) model demonstrates that if a decision maker's choices follow certain mathematical axioms, it is possible to derive Utilities - or numerical values that represent the decision maker's values - such that one alternative, which is probabilistic in nature, is preferred to another if and only if its expected utility (probability of winning multiplied by the amount to be won) is greater than that of the other alternative. One of these axioms or principles, the so-called invariance principle, is of central importance for the present studies.
Bayesian inference can be regarded as an alternative to more classical approaches to statistical testing. The method is particularly useful for the updating of base-rate probabilities by specific data. For instance, this form of analysis deals more directly with the uncertainty of population characteristics, by making reference not to a population in itself but to our beliefs of the population. In Bayesian inference, these beliefs are characterized by probability distributions and density functions.
Multiattribute utility theory (MAUT) concerns how to evaluate rationally and choose accordingly among alternatives in a multicriteria decision task. As a tool, it is used to support the decision maker and to improve his or her decision making.
Four steps are of vital importance for the analysis: structuring objectives, eliciting preferences, scoring alternatives, and finally aggregating preferences and scores (Keeney & Raiffa, 1976; von Winterfeldt & Edwards, 1986). After performing these four steps the focus of the decision maker can switch to the implementation of action.
1.2 General introduction to the present studies
According to normative decision theory there exists a principle of procedure
invariance which states that the preference order of a decision maker should remain
the same, independently of which response mode is used. For example, the decision
maker should reveal a preference indifference independently of whether he or she has to judge or decide. Nevertheless, previous research in behavioral decision making has suggested that judgments and choices yield different preference orders in both the risky and the riskless domain (Fischer & Hawkins, 1993; Goldstein, &
Einhorn, 1987; Lichtenstein, & Slovic, 1971, 1973; Slovic, Griffin, & Tversky, 1990; Slovic, & Lichtenstein, 1983; Tversky, Sattath, & Slovic, 1988;) In the latter, the prominence effect has been demonstrated.
To account for violations of procedure invariance, several explanations which emphasize distortions in the weighting of attributes have been offered. In the risky context, both Expression Theory (Goldstein & Einhorn, 1987) and Change-of- Process Theory (Mellers, Chang, Birnbaum, & Ordonez, 1992) give psychological explanations of the so-called preference reversals phenomena. In the riskless context, Contingent Weighting Theory (Tversky et al. 1988) provide a psychological account based primarily on decision-strategy compatibility. This latter account will here be subject to investigation.
The possibility is also raised that violations of invariance in the riskless context can be explained by information restructuring approaches. Two such approaches have been launched. The first one builds on Dominance Structuring Theory (Montgomery, 1983) and Differentiation and Consolidation Theory (Svenson, 1992). The general idea is that decision makers differentiate between alternatives by value and belief restructuring, by up or downgrading of alternatives according to decision rules or strategies. In the second approach, violations of invariance are assumed to be attributed to the information structure of the task which in many cases demand problem simplification (Payne, Bettman, Coupey, and Johnson, 1990).
The following questions are emphasized: What are the characteristics of making
decisions in comparison with other similar activities, such as making preference
judgments? How do situational factors affect the way decisions are made? In what
respect is decision making and judging governed by constructive processes that
take place during the process? Several of these questions will be dealt with in the
present dissertation.
2.0 Violations of Invariance
2.1 Assumptions of invariance
Expected utility theory was published by John von Neumann and Oskar Morgenstern (1947) as a theory of declining marginal utility. It was proposed as a normative theory of decision, that is, its intention was to describe how people behave if they follow certain requirements of rational decision making. According to the theory decision makers are supposed to maximize expected utility, or personal value, rather than monetary outcomes of decisions. In gamble situations, the expected value of each risky alternative is therefore equal to the probability of winning multiplied by the amount to be won.
A major objective of the theory was to provide an explicit set of assumptions, or axioms, as a basis for rational decision making. When these axioms were specified, decision researchers could use them to compare the mathematical predictions of expected utility theory with people's real behavior. A basic assumption was that people are able to express both consistent beliefs (predictive judgments) and consistent preferences (evaluative judgments). Another conjecture of the theory was that beliefs and preferences should be independent of each other, so that what you think is going to happen does not have an impact on what you would like to happen, or vice versa. Consequently, this theory of rational choice had as its main goal to describe the decisions of an idealized person. However, the theory says little about behavior per se; it is to a large extent a purely mathematical model that discusses utility theory 's relevance to optimal economic decisions.
The dominance principle is one of the axioms of rational decision making.
According to this principle, strategies which are "dominated" by other strategies should never be adopted by the decision maker. It is noteworthy that the principle only applies to situations where altematives are represented by several attributes, which can be ranked in importance order. Given these prerequisites, pairwise dominance tests of altematives can be used. The principle can in this way be used to reduce the choice-set of a decision situation, and it is therefore argued that it simplifies choice.
The so called invariance principle is another of the axioms. It stipulates that the
way options are presented should not affect the decision maker. In other words the
choices of the decision maker should be invariant with respect to either how he or
she is required to make the choice or the manner in which the options have been
presented. For example, in terms of response modes, the preference of a decision maker between two jobs should not be affected of whether the preference is to be stated in terms of a choice based on comparison, or by a contingent value judgment, that is by a pricing procedure of how much each job is worth.
According to Tversky and Kahneman (1986) the four most basic rationality principles can be ordered by their normative appeal. The invariance principle together with the dominance principle seem to be the two most fundamental principles of decision theory and are commonly regarded as normatively essential.
The status of the two other principles are more obscure: the transitivity principle could be questioned, and the cancellation principle has been rejected by many researchers. Thus, if the invariance and dominance principles are found to be invalid from a descriptive point of view, this would be most damaging to the normative theory.
2.2 Distortions in attribute weighting
One stream of research has dominated the debate on how humans make decisions. According to this stream, the nature of human judgment and decision making is emphasized as deficient. The general notion is that human capacity to make judgments and decisions is limited, leading to violations of the invariance principle (Jeans & Mann, 1977; Nisbett & Ross, 1980; Simon, 1955, 1979, 1986;
Slovic, 1972; Tversky & Kahneman, 1974; see Jungermann, 1986, for a review).
Violations of invariance can be explained in terms of the effect of different frames for the decision problem on people's decision making behavior (Payne, 1982; Tversky & Kahneman, 1981). Often these instances are referred to as violations of descriptive invariance. According to Kahneman and Tversky (1979) coding outcomes in terms of gains and losses is one of several cognitive mechanisms which people use to edit, or represent, decision problems before the options will be evaluated. As in the case of judgmental biases these operations may as well lead to violations of rationality. The errors that occur are thought of as perceptual illusions (Tversky & Kahneman, 1981). In these situations, which are frame induced, subjects evaluate problems that differ in their surface structure but share the same basic deep structure. The explanation is that decision makers tend to operate on an edited version of the decision problem.
A well-known failure of descriptive invariance was reported by McNeil, Pauker,
Sox, and Tversky (1982). They were investigating preferences between medical
treatments for lung cancer, and subjects were provided with the statistical
background for the outcomes of these treatments. Two experimental conditions
were established, one in which subjects were informed about the mortality råtes of the treatments, another in which information was presented in terms of survival råtes. The expected values of the treatments of both conditions were equivalent.
After the presentation of the problem, subjects were instructed to elicit their preferences. The results revealed a clear framing effect. The overall percentage that favored one form of therapy rose from 18% in the survival frame to 44% in the mortality frame.
Recent research has also revealed violations of the invariance principle when risk is added to mortality and survival råtes (Tversky & Kahneman, 1981). In the so-called Asian disease problem two medical programs were presented, one of which the outcome was certain, the other of which it was uncertain. Outcomes were framed differently in two conditions, that is, in terms of people saved or people lost as a result of each program. The expected values of the two programs of each condition were equivalent. Again, the results were found to violate the invariance principle. Evidently, individuals' decision frames differed in crucial ways from a formål representation of the problem.
Thus, variations in the framing of decision problems can produce systematic violations of the invariance principle, which are difficult to defend from a normative point of view. In trying to find a psychological explanation of why these violations occur, Kahneman and Tversky (1979) argued that the decision making process can be divided into two phases: an editing phase responsible for developing a decision frame and an evaluation phase during which the framed courses of actions are evaluated as a basis for choice. The operation of coding was suggested as a central process in the editing phase for the development of the decision frame.
An important issue to examine is which mechanisms could ensure the invariance of preferences (Tversky & Kahneman, 1986). It has thus been suggested that invariance would be maintained if the outcomes of the two decisions are aggregated in one kind of frame prior to evaluation, for instance Ln terms of Iives saved. The violations of invariance that have been found can be viewed as indications of that people do not spontaneously aggregate coexisting courses of actions or use one frame for the outcomes. Still, the principle of invariance could hold even if multiple perspectives are maintained. This would require that the courses of actions were separately linear in probability and monetary outcome.
However, the principle of invariance commonly fails due to the fact that the evaluation of outcomes and probabilities generally is non-linear, and because people do not use a common frame in their representations of the decisions.
Failures of invariance can therefore be described in terms of framing effects that
control the representation of options, in combination with the nonlinearities of value and belief.
The emphasis on editing or representation of decision problems has added something new to the research. Previously, scientist's major objective was concentrated to describe final judgments and evaluations on behalf of the earlier stages of the decision making process.
Another important form of violation of the invariance principle is termed procedure variance. Situations belonging to this category are task induced, and the preference ordering is found to differ depending on the mode of response required by the task. A generally accepted explanation of violations of procedure invariance is that the response mode affects how the attributes are weighted. This issue is highlighted in the research on so-called preference reversals (e.g., Lichtenstein and Slovic, 1971, 1973; Slovic et al, 1990; Slovic & Lichtenstein, 1983, Tversky et al., 1988; see Payne, Bettman, & Johnson, 1992, for a review). In normative theories it is assumed that specific empirical procedures like choice or pricing should give the same results. A preference reversal occurs whenever an individual prefers one alternative in one procedure but shows the opposite preference order in another.
There is much evidence showing that the price ordering of risky prospects is systematically different from the choice ordering. For example, when subjects have to indicate which bets they prefer and how much they would be willing to sell the bets for, they often choose the high-probability option but indicate the highest selling price for the high-payoff option. The explanation of the phenomenon attributable to Lichtenstein and Slovic (1971, 1973) is that subjects in judgments set minimum selling prices by a process of anchoring and adjustment. As an example, subjects who find a gamble attractive are believed to anchor on the amount they stånd to win and then adjust downward for the amount and probability to lose. This is not the case in choice modes. Choice, it is argued, provides a freedom to use any decision strategy, whereas judgments often provide natural starting points, due to scale compatibility. In pricing, for instance, the natural starting point is provided in the gamble by the amount to win. Nevertheless, as many researchers know, probabilities are difficult for subjects to translate into monetary units. As a result, the adjustment downwards becomes insufficient, leading to a preference reversal. Because of this, an overpricing effect of gambles is established. However, it has been pointed o ut that this kind of adjustment is insufficient as an explanation (Tversky & Kahneman, 1974).
Other competing accounts of the preference reversals phenomena have been
proposed by Goldstein and Einhorn, (1987) and Mellers et al. (1992). In
Expression Theory, Goldstein and Einhorn (1987) state that previous research has
confounded judgment versus choice with the response scale used to express preference. For instance, prices have been expressed in dollars, ratings on rating scales. Instead, these factors are suggested to be crossed. They attribute the reversals found in the conditions to changes in the mapping from a gamble's components to the response. This transformation is assumed to differ predictably for each gamble and for each response mode.
In Change-of-Process theory (Mellers et al. 1992), it is assumed that the stimulus context also has the ability to influence preference orders. It is proposed that the inclusion of certain stimuli "steer" subjects away from one decision strategy and toward another. Different preference orders are assumed when subjects use different procedures to evaluate the quality of gambles. Hence, people are supposed to use different decision strategies in different tasks with the same scales. Such tasks can be the rating of attractiveness and risk, pricing, and the judging of strength-of-preferences.
In more recent research it has been established that the majority of reversals are violations of invariance, caused by over-pricing of the low-probability high-payoff bets (Tversky, Slovic, & Kahneman, 1990). Thus, intransitivity and failure of independence only account for a small proportion of the reversals (10%). The overpricing effect can be attributed to an effect of scale compatibility: When people are asked to set a price for how valuable the bet is, they look at how large the potential payoffs are. Payoffs are weighted more heavily in pricing than in choice because prices and payoffs are expressed in the same units. On the other hand, when people are asked to choose between two bets, they pay particular attention to the probability of winning. Thus, choosing between a pair of gambles seems to involve different psychological processes than bidding for each one separately.
Response-mode biases have also been found when subjects are asked to evaluate pairs of decision alternatives whose consequences are described by two attributes. Tversky et al. (1990) and Slovic et al. (1988) demonstrated a judgment- choice discrepancy, in the form of a riskless preference reversal, in such a case.
One of the attributes was selected to be predominant or prominent. In choice tasks subjects placed more weight on this attribute than they did in matching tasks in which they were required to make the two options equally attractive.
The compatibility hypothesis was proposed to account for the prominence effect
(Tversky et al., 1988; Slovic et al., 1990; Fisher & Hawkins, 1993; Hawkins,
1994). It is based on the pioneering work of Slovic (1975). In this hypothesis the
effect reflects a general principle of compatibility according to which the
processing of input (e.g., attributes describing options in a judgment or choice
task) depends on how compatible it is with the output (i.e. subjects' responses).
Tversky et al. (1988) argued that both risky and riskless preference reversals are governed by stimulus-response compatibility. Identical components on both the stimulus and response side enhance compatibility. Such components are the use of the same scale units (e.g., grades, ranks), the direction of relations (e.g. whether the correlation between input and output variables is positive or negative), and the numerical correspondence (e.g., similarity between the input and output). The use of similar scale units have been referred to as scale compatibility by Fischer and Hawkins (1993) and Hawkins (1994). According to Slovic et al. (1990) the theoretical definition of scale compatibility is somewhat loose but its implications seem unambiguous. However, scale compatibility alone cannot account for the prominence effect (Hershey & Schoemaker, 1985; Schkade & Johnson, 1989;
Slovic et al. 1990). It may account for variations in strength of preference found in different judgment modes, but in judgment-choice comparisons this type of compatibility can yield predictions directly opposed to the prominence effect (Fischer & Hawkins, 1993). For instance, if dollar is the prominent attribute, the scale-compatibility hypothesis implies that people will attach greater weight to money in dollar-matching tasks than in choice.
Another suggestion is that the prominence effect occurs because choice and matching tasks evoke different types of decision strategies giving different weight to the prominent attribute (Tversky et al. 1988). The qualitative response in choice is regarded as compatible with a qualitative lexicographic decision rule which renders quantitative weighting of attributes unnecessary. In contrast, quantitative judgments are compatible with a quantitative weighted additive rule. This form of compatibility is referred to as strategy compatibility (Fischer & Hawkins, 1993;
Hawkins 1994). The cognitive basis of this suggestion is twofold. First, as in the former case of scale compatibility, noncompatibility is regarded as requiring additional mental operations which subjects avoid. Second, a response mode is seen as priming the focus of attention on those compatible features of the input that are compatible with the required output.
In conclusion, the theories provide cognitive explanations of violations of the
invariance principle. For instance, observed response mood effects serve as a
foundation for cognitive hypotheses, predicting which strategy is to be used in
what response mode. A widely used perspective in this stream of research is that
information processing capacity is limited, which leads to violations of rationality
principles. The causes for this is often traced to the use of heuristics, to the
representation of the problem, or to motivational factors. Normative models or the
objective reality are not seldom used as rational yardsticks. A general growing
consensus is that preferences and beliefs are constructive in nature. This means that
preferences for and beliefs about objects or events of any complexity are often constracted - not merely revealed- in the generation of a response to a judgment or choice task. This argument will also be highlighted in other directions of research to be reviewed.
2.3 Cost-benefit theories
A general idea put forward in cost-benefit theories is that decisions which violate the invariance principle nevertheless can be described as rational if the cognitive costs are taken into account. Hence, the decision costs must be weighted against the potential benefits resulting from the application of a decision strategy.
This may lead to violations of invariance as it is stated in the subjective expected utility model (SEU). However, such violations are, according to the theories, perfectly rational (Beach, 1990; Beach & Mitchell, 1978, 1987; Einhorn &
Hogarth, 1981). The selection of which decision rule or strategy is to be used (i.e., SEU, maximizing, satisficing, elimination-by-aspects, etc.) in a specific situation depends both of the decision maker's desire to make the best decision and his or her attitude towards investing time and effort in making a decision. Which strategy is perceived as superior in maximum net gain is the one being selected. Violations of invariance in the "classical sense" is thus something that the decision maker anticipates and tolerates. Which are then the benefits and costs for resolving the confiict inherent in choice, that is, which are the pros and cons of the engagement in the required mental effort? On the benefit side, one can put forward such gains as environmental control, clarifications of goals and preferences, creation of the habit of thinking, reformulation of problems which lead to discovering of new alternatives, and seeking ways of information that might resolve the confiict in choice. These five benefits are quite general in nature and apply across all kinds of consequential choice. It is interesting to note that the trade-offs of these benefits must be viewed from a temporal perspective. Short-term investments in terms of mental effort related to thinking and reasoning has long-term benefits in the sense of enhanced action control and improved capability of handling choice situations adequately. On the cost side, thinking and reasoning can be distressful in a state of preference uncertainty when you do not know what you want. A second cost of thinking is that the trade-offs in the decision situation are made visible. Third, since humans have a limited capacity to process information, there are costs due to acquiring, processing, and outputting information (Hogarth, 1987).
To be able to get insight into the costs and benefits of thinking, and to identify
the combinations of decision rules being used by decision makers, Payne (1976)
suggested that making inferences from behavior is not sufficient. Instead, he proposed that the study of acquisition and search of information should be emphasized as the prescribed research methods in the study of decision processes.
In this respect, he was heavily influenced by models of human problem solving at that time (Newell & Simon, 1972). Another influence came from the development of the so-called information board technique. In a typical information-board experiment, matrices of alternatives by attributes are presented on a computer screen. Subjects are searching for information before stating their preference.
The studies of Ericsson and Simon (1980, 1984) dealing with verbal reports as data provided a source of inspiration for the development of another widely used technique, the protocol analysis. The major objective in studies involving both information board and protocol analysis techniques has been to identify whether decision processes follow compensatory or non-compensatory strategies
1(see Ford et al. 1989, for a review). In fact the study of violations of the invariance principle, from the strategy identification point of view, has been one of the major issues. For instance, the effects of response mode on decision making strategies were investigated by Billings and Scherer (1988). They found that there were more information searched/ interdimensional processed in judgment response modes than in choice. They also found a more constant amount of information searched across alternatives in judgment than in choice.
In their attempt to identify the strategies used in different response modes in risky contexts, Schkade and Johnson (1989) studied cognitive processes in preference reversal gambles. In the first two experiments, they analyzed pricing-
^Examples of compensatory decision rules or strategies are the additive utility and the additive difference rules. In these rules it is assumed that each dimension can be measured on an interval or ratio scale and be given a weight that corresponds to its importance. Each altemative is then evaluated by summing the weighted values on each dimension. In the additive difference rule the decision maker is assumed to evaluate the differences between the alternatives on a dimension by dimension basis, that is, the difference between the alternatives on each attribute is calculated and summed. The decision maker then choses the altemative with the greatest aggregated difference.
Among the non-compensatory strategies, four can be identified as the most frequently studied: the
conjunctive, disjunctive, lexicographic, and elimination-by-aspects (EBA) strategies. In the
conjunctive strategy the decision maker defmes certain cut-off points or thresholds on the
dimensions, and if an altemative on any dimension fails to reach any of these thresholds, it is
rejected. A disjunctive model is on the other hand characterized by that a decision maker can
accept a low score of an altemative on one dimension, provided that it has a high score on
another dimension. In the lexicographic strategy, the decision maker makes an importance
ordering of the dimensions, and then makes a choice on the basis of the most important
dimension. The elimination-by-aspects (EBA) strategy is related to both the lexicographic and
the conjunctive strategy. It is assumed in it that alternatives consist of a set of aspects, and that
alternatives that do not exceed the required point is rejected, beginning with the most important
dimension. All the "failing" aspects are eliminated dimension-wise, until there only is one
surviving altemative.
choice and pricing-rating reversals. An important question was the role of anchoring and adjustment for the preference reversal effect. Attention to attributes and total time was used as dependent measures. The process analysis carried out in the first experiment (pricing-choice) revealed that pricing generally took longer time than choice. Another finding was that more attention was directed towards the pay-off information in pricing than in choice. Furthermore, less attention was directed towards the probability råtes in pricing than in choice. The process analysis carried out in the second experiment (pricing-rating) showed that the starting points were affected by the compatible gamble elements in both pricing and ratings. As a rational for this, the anchoring and adjustment principle was suggested.
Westenberg and Koele (1990, 1992) investigated the influence of response modes on decision strategies. Their general hypothesis was that the type of response required in a decision situation influences which decision strategy is selected. In both their studies subjects to a higher extent were found to use compensatory strategies as a function of the increasing number of categories (or scale-points) in the decision situation. In line with Billings and Scherer (1988) qualitative response modes tended to induce more non-compensatory strategies than judgments. Furthermore, an interesting difference was found between the two qualitative response modes of selecting and rejecting. Both response modes were found to induce non-compensatory strategies, but compensatory strategies were more frequent in the rejection than in selection mode.
This methodological approach was used in the present studies to validate one of the explanations of riskless preference reversals; the strategy compatibility hypothesis (Hypothesis Hj). Hence, the validity of the hypothesis will be tested with the use of both input and output data together with cognitive accounts in line with the information-search tradition (Russo, Johnson, & Stephens, 1989; Maule, 1992; Westenberg, 1992).
In conclusion, cost-benefit explanations of violations of invariance state that
people have available or can generate different decision strategies for making
choices or judgments. It is also assumed that the strategies differ in expected costs
and benefits. To be able to select a strategy, the decision maker must take in to
consideration the anticipated costs and benefits of each strategy, given the specific
task environment. In this stream of research, process tracing is typically carried
out with information board techniques or protocol analysis (see also Johnson et al.,
1988, for a review).
2.4 Restructuring theories
In Dominance Structure Theory (DST) (Montgomery, 1983) it is argued that the most cost-effective way for people to resolve the conflict inherent in choices is to find a cognitive structure which yields dominance for one alternative.
Dominance structuring is seen as a principle for facilitating the information processing in the decision situation. The main idea is that people try to structure and restructure their representation of a decision problem so that one alternative always appears to be the best. The decision process is hence seen as a search for a dominance structure, that is, as a cognitive process where all possible disadvantages of a promising alternative are eliminated or neutralized. This alternative then appears as the best one, since it can be maintained that it dominates the other ones. It is important to mention that if the decision maker fails to find a dominance structure for a promising alternative, it is always possible to return to the previous phase and carry on the search for dominance (see also Dahlstrand &
Montgomery, 1989, Montgomery & Hemlin, 1991, Montgomery & Svenson, 1989 for empirical tests of the theory).
A more recent contribution is Differentiation and Consolidation Theory (DCT) (Svenson, 1992). In this theory, decision making is conceived of as a process in which one alternative gradually is differentiated from another until the degree of differentiation is enough for a decision. Thus, it is not sufficient to choose an alternative that is better. Rather, the preferred alternative should be so much better than the nonpreferred alternative that it remains better even under unfavorable post-decision conditions
2(see also Svenson & Benthorn, 1992, and Svenson &
Malmsten, 1992, for empirical tests of the theory). However, the present studies are primarily based on the theoretical foundations of DST.
The primary aim of DST is to describe how choices or decisions actually are made. To be able to understand how the theory can explain violations of invariance, such as the prominence effect reported by Tversky et al. (1988), it is motivated to parallel the different steps of one of these models with the steps prescribed by a theory dealing with judgments in detail (see Tables 1 and 2). As mentioned earlier, multi-attribute utility theory (MAUT) provides such an account.
The reason for this is to give insight into (i) why choices are more unpredictable from linear weighting models than judgments, and (ii) why the predominant
2
Svenson (1992) actually postulates three types of structural differentiation modes or ways of changing the representation which all - at least in part - build on value and belief restructuring.
The modes are labeled differentiation through attractiveness restructuring, differentiation
through attribute importance restructuring, and differentiation through facts restructuring (see
also Fishhoff, 1975, and Hogarth & Einhorn, 1985).
attribute looms larger in choice than in matching judgments. Among several attribute weighting models, the so-called direct weight elicitation method (DWT) (Edwards, 1977) was selected for this purpose, on the basis of its simplicity and theoretical soundness (see Borcherding, Schmeer, & Weber, 1993, for a review).
Table 1. The phases of DST explaining choices.
Pre-editing phase. All unnecessary attributes and alternatives are deleted from further processing.
Finding a promising alternative phase. This phase implies that the decision maker detects an alternative with attractive attributes or at least an altemative with attributes that encourage great expectations of how such an altemative is supposed to be.
The dominance testing phase. This is a sort of everyday sensitivity analysis in which the decision maker tests if there are any disadvantages connected to the promising alternative. If no disadvantages are found, the promising alternative is chosen. The test can be carried out more or less systematic, depending on the felt pressure to choose the promising alternative.
The dominance structuring phase. This phase aims at "repairing" the deviations from dominance which have been localized in the previous phase. This aim can be completed by using different kinds of operations, such as bolstering of the advantages or de-emphasizing of the disadvantages of the promising alternative.
But two other operations, cancellation or collapsing may also be used. Cancellation implies that the decision maker makes trade-offs between advantages and disadvantages of the alternative. This operation is often used when there exists a natural connection between an advantage and a disadvantage. Collapsing on the other hand implies that two or more attributes are collapsed into a superior attribute, which often is more comprehensive (see also Huber, 1989).
The search for a dominance structure is according to Montgomery (1983)
supposed to go through four phases, that is pre-editing, finding a promising
alternative, dominance testing, and dominance structuring. Edwards (1977) also
presents four phases in his model, that is structuring the problem, determining the
importance of dimensions, measuring alternatives on the dimensions, and choice.
Attempts have been made to test DST as a psychological account of violations of procedure invariance. Lindberg, Gärling, and Montgomery (1989), found it plausible that dominance structuring could be the answer to the worse fit of the predictions of multi-attribute utility theory (MAUT) to choices than to preference ratings.
Table 2. The Phases of DWT explaining judgments
Structuring the problem phase. This phase includes the following four steps: (1) The decision maker(s) is identified. (2) The decision is identified. (3) The alternatives to be evaluated are identified. (4) The dimensions on which the alternatives are to be evaluated are identified.
Determining the importance of dimensions phase. This phase includes the following three steps: (1) The dimensions are rank-ordered in terms of their importance. (2) The rankings are translated into ratings. (3) The ratings are converted to numbers that sum to 1.
Measuring alternatives on the dimensions phase. This phase includes the following two steps: (1) The alternatives are measured on each of the dimensions.
(2) The overall worth of each alternative is calculated by summing each alternativens scores on the dimensions which were weighted earlier.
Choice phase. The previous phase yields a list of alternatives accompanied by their measures of relative worth. The normative rule is to choose the alternative with the largest assessment of worth.
As described earlier, the decision maker is according to the theory assumed to modify his or her beliefs/evaluations related to attributes and attribute levels in such a way that a favored alternative stånds out as dominant. Lindberg et al.
(1989) thus argued that such changes in belief-value structures in the choice process should make the outcome of the choice difficult to predict based on previously performed belief evaluations as input data for the multiattribute utility predictions. The results were in line with an interpretation which favored the idea of simplifying heuristics used in the dominance structuring of the choice task.
These heuristics were not used in the judgment task which led to the better fit.
As noted above, Tversky et al. (1990) and Slovic et al. (1988) demonstrated a judgment-choice discrepancy in the form of a riskless preference reversal in the case where one of the ättributes were selected to be predominant or prominent. In choice tasks subjects placed more weight on this attribute than they did in matching tasks in which they were required to make the two options equally attractive. The phenomenon was termed the prominence effect. The rational provided was that the effect occurs because choice and matching tasks evoke different types of decision strategies giving different weight to the prominent attribute. The qualitative response in choice is regarded as compatible with a qualitative decision rule which renders quantitative weighting of ättributes unnecessary. In contrast, quantitative judgments are compatible with a quantitative weighting rule.
If there are one prominent and one nonprominent attribute, as a result of the restructuring process the former will have more influence since the differences on that attribute are enlarged relative to the other. More precisely, subjects may modify their beliefs or values in such a way that there will be a larger discrepancy between the options on the more important attribute than on the less important attribute. Both the importance order of the ättributes and the differences between the alternatives on the ättributes will then speak in favour of a preference in line with the prominent attribute. The above mentioned modifications of beliefs or values are thought to take place primarily in choice which is characterized by an ability to resolve conflicts between alternatives. It is assumed that reasons or motives guide the modifications of values and beliefs. Therefore, Montgomery (1983, 1989) assumed that the importance of a decision is directly related to the degree of restructuring.
A hypothesis (Hj) which will be tested in the studies is that the prominence effect can be explained by Montgomery's (1983, 1989) theory of dominance structuring in decision making. It will be referred to as the value-belief restructuring hypothesis (or sometimes just the restructuring hypothesis).
Another perspective on restructuring is taken by Payne, Bettman, Coupey, and
Johnson (1992). They suggest that one very important analytic tool for the
understanding of the contingency of preferences on task demands is the
restructuring of the input data. According to Payne et al. (1992) restructuring is
thought to occur in the editing phase, which is earlier than what Montgomery
(1983) suggested. Examples of restructuring operations are information
transformations (e.g., rounding off, standardizing, or performing calculations),
rearranging information (e.g., changing the order of the alternatives or ättributes)
or eliminating information (see also Russo, 1977, and Ranyard, 1989). The aim of
restructuring seems to be to make problems more manageable. With the help of transforming, rearranging, or eliminating information, the decision maker can use a processing strategy and obtain an acceptable amount of accuracy and cognitive effort in the task situation. That same strategy could have been too difficult to use before restructuring. A mode of differentiation proposed by Svenson (1992) in DCT encompasses this mechanism
3.
The choice problems used by Payne et al. (1992) were either well-structured or poorly structured and information was presented either simultaneously or sequentially. The characteristics of a well-structured problem was that all information for an attribute was expressed in the same units, and information was presented in the same order within each alternative for any given attribute. Poorly structured problems were characterized by the fact that the information within the same attribute had different units, and that the information could appear in a different order within each alternative for any given attribute. Payne et al. (1992) let subjects take notes while processing, and those verbal protocols were then coded with respect to specific restructuring operations. These operations were used by the subjects to create different forms of helpful matrices in the poorly structured problems. To arrive at matrix representations, subjects used transformations, calculations, and rearranging operations. Subjects who made notes were then compared with others who did not It was found that subjects who restructured (in the sense that they made notes) to a higher extent were using alternative-based strategies when processing the restructured material. The conclusion reached by Payne et al. (1992) was that subjects use restructuring as a means of mental effort investment to be able to låter use a more accurate strategy with a reasonable amount of effort. They found that restructuring takes place in the early stages of the decision process, but this finding might be due to an artifact produced by the methodology of taking notes. Hence, restructuring is thought to
3
In Differentiation and Consolodition Theory (Svenson, 1992), the fourth mode of differentiation
is based on heuristics taking place in the editing phase. Differentiation through decision problem
restructuring implies that real life decisions can be more or less well-defined, as stated earlier. In
the latter case people might want to create decision alternatives all by themselves, due to for
instance perceived uncertainty, task complexity, or social conflict. Svenson (1992) illustrates this
with a decision between manufacturers of prefabricated homes. A decision like this may involve
much mental effort in composing alternatives from the different manufacturers, due to task
complexity. Uncertainty may as well lead to decision problem restructuring. New and unfamiliar
decision situations are therefore often difficult to differentiate on the alternative level. In
situations like this, people often import a new reference alternative without seriously considering
it in the decision situation. This might lead to an increased differentiation on the process level
(see also Payne, 1982; Shafir, Simonson, & Tversky, 1993; Tyszka, 1983; and Wedell, 1991).
have the ability of occurring at any time in the decision process, although it mostly occurs in the early stages.
Instead of compatibility between qualitative/quantitative response mode and decision rule (Tversky et al. 1988), it rrtay be assumed that the required output from subjects needs to be compatible with the structure of information in input. In a matching task there is an agreement between input and the required output.
Subjects are required to match one value difference (required output) to another difference which is given in the task (input). Hence, there exists a dimensional compatibility between input and output. If the difference between attribute levels of the prominent attribute serves as input, then the difference between the levels on the nonprominent attribute serves as the required output, and vice versa. This form of compatibility involves transforming and rearranging of the information by subjects in the editing phase, since both differences always have to be taken into consideration.
In choices and preference ratings on the other hand, both differences serve as input. For example, in a two alternatives x two attributes choice task the input Ls built on four pieces of information (the attribute levels of each alternative), whereas the output corresponds to a preference order between alternatives. Here, there is no dimensional compatibility between input and output. Subjects therefore do not transform and rearrange the information to the same extent, simply because they are not forced to do it. The information structure of these tasks gives them the opportunity to select a lexicographic strategy.
Hence, whereas choice as a qualitative response mode is distinguished from judgments, the idea is offered that matching judgments can be seen as distinct from preference ratings and choices in that subjects have to evaluate one value difference relative to another to carry out the task. On the basis of this reasoning, a prominence effect is also expectéd for preference ratings, due to the assumption that information structure compatibility is salient for the selection of strategy. The crucial fact for the strategy being used is the dimensional compatibility between the input and output and not (primarily) the metric levels of the same.
The hypothesis (H3) that restructuring in the editing phase can explain the prominence effect is investigated in the present studies. It is henceforth referred to as the structure compatibility hypothesis.
There are probably connections between restructuring in the editing phase and
value-belief restructuring. Henceforth, it is generally assumed that the information
structure in input of different response modes in terms of presentation mode of the
alternatives (sequential, simultaneous) has an effect on the belief-value
restructuring låter to come. Therefore, it is assumed that if the information
structure is characterized by sequential presentation ofalternatives, this will lead to a decrease in value-belief restructuring, and a reduction of the prominence effect.
If, on the other hand, the information structure is characterized by simultaneous presentation of alternatives, then this will lead to an increase of the value-belief restructuring, and to an enhanced prominence effect (see also Birnbaum, 1992).
To conclude, in cognitive restructuring theories, violations of the procedure invariance principle is interpreted as resulting from constructive processing in choice. In the phase of restructuring a tentatively promising alternative is differentiated from another/others. Because of this the prominent attribute looms larger in choice than in other response modes, for instance in judgments. The prominence effect in choice is therefore assumed to be acquired by different forms of restructuring operations in which changes of the representations take place.
Researchers which adopt this perspective challenge the view that human choice is cognitively deficient
4. Different arguments can be used in support of this standpoint. First, it can be argued that an important parameter has been neglected which is the cost of a decision strategy. Second, decisions must be viewed as parts of a continuous constructive process. They are nevertheless often seen as discrete events. Third, it is argued that very little attention has been given to the internal structural representation of the problem. A central issue in all three arguments is what conditions humans reveal in which kind of behavior. The issue of human behavioral efficiency from a general point of view is thus of less interest.
4
